Soft-Sensor of the Key Water Quality Indicators in the Wastewater Treatment Process based on Improved Mamba

With the ongoing global transformation in algorithms and advances in hardware, the vertical application of artificial intelligence (AI) across various domains has garnered significant attention. In recent years, AI models represented by deep time series models have rapidly developed due to their ability to effectively address nonlinearity and dynamics. In wastewater treatment processes (WWTP), the coexistence of nonlinearity and dynamics presents a critical challenge to soft-sensor modeling, leading many scholars to make remarkable contributions based on deep time series models. However, there is still room for improvement in the mechanisms of these models, and existing methods lack sufficient capability to extract multi-scale spatiotemporal features, including nonlinearity and dynamics. This limitation affects the efficiency and performance of current models, hindering their vertical application. To address these issues, Improved Mamba (ImMamba) method is proposed based on the Mamba Network for modeling soft-sensor of the key indicators in WWTP. The proposed method leveragesWavelet Transform Module to improve multi-scale feature extraction and employs an inverted Data Embedding Module to capture spatial features, thereby enhancing the representation of multi-scale spatiotemporal features. Additionally, leveraging the advantages of the Mamba architecture, the ImMamba model effectively addresses the challenges faced by previous models in application by enhancing training efficiency and capturing long-term dependencies, thereby promoting the vertical application of AI models in the field of wastewater treatment. The effectiveness and efficiency of the proposed method are validated through comparisons with state-of-the-art algorithms.